Abstract:In order to solve the problem that wind power is hard to predict due to its characteristics such as large fluctuation and strong randomness, a ultra-short-term wind power prediction model for wind farms was proposed based on fuzzy C-means (FCM), which selects similar daily and salp swarm algorithm to optimize the extreme learning machine (SSA–ELM). Firstly, the FCM data clustering method was used to select similar days with higher correlation with the predicted days, of which the historical wind speed, temperature, wind direction and other climatic factors that are highly correlated with wind power formed a multi-input sample set. Secondly, the network parameters were determined in the training process through the training set, and the input weight matrix and hidden layer deviation of the extremely learning machine were optimized to improve the adaptability and accuracy of the prediction model by using the salp swarm algorithm in an iterative process. Finally, according to the ultra-short-term wind power interconnection related regulations, using the actual data of a wind farm in Henan province from ultra-short-term prediction based on similar days, three aspects of the four seasons and rolling prediction error of the representative simulation experiment, and the extreme learning machine (ELM) and BP neural network model were analyzed. The results show that the proposed model has faster convergence speed and higher prediction precision. It is demonstrated that the proposed ultra-short-term wind power prediction model based on FCM and SSA–ELM has good tracking and generalization ability.